Medical Visual Question Answering on VQA-RAD n=350 (test)
77.7LLM-Judge AccuracyGemini-3.5-Flash
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-3.5-FlashEvaluation Protocol=zero-shot, Model Category=frontier VLMs2026.07 | 77.7 | |
| Claude-Sonnet-4.6Evaluation Protocol=zero-shot, Model Category=frontier VLMs2026.07 | 65.4 | |
| DiffusionGemmaEvaluation Protocol=per-dataset finetuned, Backbone=diff2026.07 | 64.9 | |
| Gemma-4Evaluation Protocol=per-dataset finetuned, Backbone=AR2026.07 | 64.9 | |
| DiffusionGemmaEvaluation Protocol=base (zero-shot), Backbone=diff2026.07 | 61.4 | |
| GPT-4.1-miniEvaluation Protocol=zero-shot, Model Category=frontier VLMs2026.07 | 57.1 | |
| Gemma-4Evaluation Protocol=base (zero-shot), Backbone=AR2026.07 | 52.3 |